From: kettenis <kettenis@openbsd.org>
Date: Fri, 25 Jul 2008 19:37:16 +0000 (+0000)
Subject: Work around a problem on the t5120 where top(1) would lock up the system for
X-Git-Url: http://artulab.com/gitweb/?a=commitdiff_plain;h=ef642599c078b0a2bd2f1c29bb1be9b3f317b614;p=openbsd

Work around a problem on the t5120 where top(1) would lock up the system for
several minutes by making pmap_resident_count() return 0 for the kernel pmap.

discussed with miod@
---

diff --git a/sys/arch/sparc64/sparc64/pmap.c b/sys/arch/sparc64/sparc64/pmap.c
index 46be8e6a088..1bc33e07f67 100644
--- a/sys/arch/sparc64/sparc64/pmap.c
+++ b/sys/arch/sparc64/sparc64/pmap.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.c,v 1.65 2008/07/05 21:20:48 kettenis Exp $	*/
+/*	$OpenBSD: pmap.c,v 1.66 2008/07/25 19:37:16 kettenis Exp $	*/
 /*	$NetBSD: pmap.c,v 1.107 2001/08/31 16:47:41 eeh Exp $	*/
 #undef	NO_VCACHE /* Don't forget the locked TLB in dostart */
 /*
@@ -3192,6 +3192,15 @@ pmap_count_res(pm)
 	paddr_t *pdir, *ptbl;
 	/* Almost the same as pmap_collect() */
 
+	/*
+	 * XXX On the SPARC Enterprise T5120, counting the number of
+	 * pages in the kernel pmap is ridiculously slow.  Since ps(1)
+	 * doesn't use the information for P_SYSTEM processes, we may
+	 * as well skip the counting and return zero immediately.
+	 */
+	if (pm == pmap_kernel())
+		return 0;
+
 	/* Don't want one of these pages reused while we're reading it. */
 	s = splvm();
 	simple_lock(&pm->pm_lock);